The increasing sophistication of artificial intelligence in generating human-like text presents significant ethical, social, and legal challenges. In response, this study introduces a novel machine learning-based ensemble model to detect and classify AI-generated and human-written text. The model integrates K-Nearest Neighbors, Logistic Regression, Gaussian Naive Bayes, and Random Forest classifiers using a soft voting mechanism for robust classification—the Microsoft MiniLM model utilized to extract the features from the data. The PCA and t-SNE method utilized to reduce the dimensionality of data and select the relevant features and visualize the patterns. Extensive experimentation on a balanced dataset of 50,000 text samples shows the ensemble model’s superior performance, obtaining 97.34% accuracy, outperforming existing models. The proposed approach addresses the growing need for reliable detection systems to mitigate misinformation and misuse of AI-generated content.